I have been in touch, on and off, with John Chambers for many years, since he worked in the BBC Research Department at Kingswood Warren. When he took early retirement due to the cut-backs, it was no surprise to find he had landed a job with the National Physical Laboratory at Teddington as their "Head of Time and Frequency Services". At the beginning of 1999, after some difficult questions from colleagues about the recent leap-second, I emailed John and asked if he could explain a little of the background to why we had leap-seconds. Below is his reply:
Subject: You asked for it!
Date: 13 January 1999 09:39
I'm happy (within limits) to try to explain, it helps me practise for when I'm doing it 'for real' to schoolchildren, worthy aged professors, etc.
Until mid 1950s we had 'elastic seconds', although we did not realise this. We were dividing (mean solar) days into 86400 seconds, not realising that the length of the day was varying in a range of about 3 ms. By the way, length of (mean solar) day is 'worked back' from length of the sidereal day (as measured by star transits, or, these days, by pulsar positions) which measures the Earth's period of spin against 'space' rather than the sun, thus avoiding the 'mean solar' bit due to elliptic orbit and tilt of Earth's axis which makes sundials only correct four times a year and up to 16 minutes out at other times.
In 1949 the NBS in Washington produced an ammonia-based atomic frequency standard good to about 1 part in 10^7 - not quite as good as the best quartz or pendulum clocks. In 1955 at NPL Louis Essen produced the first Caesium-based atomic clock good to about 1 part in 10^9 - enough to show that the Earth's rotation period was not constant.
Scientists (and everyone else) like their standards of measurement to be (1) constant and (2) easily reproduced (e.g. freezing and boiling points of water for temperature).
As a last-ditch attempt to keep control of the second, and avoid it falling into the hands of the scientists, the astronomers realised that the length of the (tropical) year (measured from, for example, when the sun 'crosses the equator northbound' at the spring equinox) was more stable than the Earth's rotation, so in 1956 the second (which previously, amazingly, had not really been defined at all because 'everyone knew what it was') was defined as
the fraction 1/31 556 925.9747 of the tropical year for 1900 January 0 at 12 hours Ephemeris time.
(perhaps not quite the same thing as 1/86 400 of the mean solar day for 1900).
Astronomers are more concerned with motion of heavenly bodies than where the sun is in the sky at Greenwich and they used to write all their equations based on t=0 at year 1900. I guess that is why this year was chosen. But, of course, this definition fails the above '(2) easily reproduced' test rather badly! By the way, they currently write all their equations (e.g. for the forthcoming eclipse) based on t=0 at year 2000.
This mess didn't last long and in 1967 (amazingly quick for international agreement) the second became -
the duration of 9 192 631 770 periods of the radiation corresponding to the transition between the two hyperfine levels of the ground state of the caesium 133 atom.
In 1956 an international atomic time scale (TAI) was started, counting these new super-regular atomic seconds, for scientific purposes. The timescale used by astronomers used to be known as Ephemeris time and is now known as Terrestrial Dynamic Time (TDT). You get it by adding 32.184 s to TAI (International Atomic Time).
Now the punch line, which you never find written in the books but which (to me at least) is blindingly obvious (Chambers' first law of time?) -
No two independent clocks, even when synchronised at a given point, will keep in step for ever - they are bound to drift slowly apart.
Or - to put it another way - there are no gear wheels between the Caesium atom and the solar system.
So there was a very messy period in the 1960s when step adjustments of 0.1 s or 0.05 s were made from time to time, and annual 'frequency offsets' for standard-frequency broadcasts (e.g. - 130 parts in 10^10) were declared, to synthesise a 'public' time scale from the 'regular' atomic time (that was being maintained and shared by more and more laboratories and observatories) that would keep in step with the 'traditional' time. These adjustments were primarily made for the benefit of traditional navigators - the Earth rotates at about 500 m/s at the equator (about 285 m/s at our latitude), so an error of one second can mean quite an error in position.
This situation where, for example, there were 86 400.002 54 seconds pulses of MSF in a 'day', and where laboratories were being asked to 'readjust' their 'working standard seconds' every year, was intolerable and confusing.
In 1972 a new Coordinated Universal Time (UTC - deliberately not right in any major language!) scale was defined (the earlier mess was also known as Coordinated Universal Time, Universal Time being (and is still) what we know as Greenwich Mean Time).
From 1972 January 1 all time signals followed UTC. That is when the sixth pip of the GTS was lengthened. An adjustment of about -0.108 s was made to UTC to make it exactly 10 s behind TAI. UTC seconds became identical in length to atomic seconds (no more precision frequency offsets) and all future corrections were to be made in whole seconds (leap seconds). These would be inserted (or deleted if ever necessary) at the end of December or end of (UTC) June, or, in emergency, end of March or September. You can see on our web site the cumulative list of leap seconds so far. For almost all of the time (except, as discussed, perhaps later this year for a brief period) UTC has been 'faster' than the Earth's rotation so we've needed positive leap seconds, 61-second minutes, to let UTC fall behind and start catching up again.
There is a table (probably in the IERS website) giving mean length of day (in atomic seconds, of course) back to the 1600s. This has been done by reconciling contemporary observations of known astronomical events, such as eclipses and occultations. 1900 might seem to have been a 'naff' year to choose, but any choice would give problems at some time or another. As indicated above, it was chosen to give continuity with the definition selected by the astronomers.
In the long term (hundreds of years) the Earth is slowing down due to tidal friction. But short term, it's all over the place due to weather and movements inside the Earth's crust. It irritates me to hear 'We need leap seconds because the Earth is slowing down'. We would also need leap seconds if the Earth were speeding up. We also need leap seconds (at a steady rate) even if the Earth is spinning at a steady rate. It is the 'rate at which we need leap seconds' that increases if the Earth is slowing down.
IERS publishes via its Bulletin D the 'DUT1' figure, which changes at 0000 UTC on Thursdays when necessary (and at leap seconds), which gives the difference between UT (GMT) and UTC rounded to one-tenth of a second, to make UT more accessible to traditional navigators. This figure is broadcast as part of the MSF signal (currently seven double pulses in seconds 01 to 07 showing that DUT1 is +7, UT (GMT) is about 0.7 s ahead of UTC. Just before the last leap second DUT1 was -3. This figure is also signalled in the format of our Truetime telephone time service.
The GPS system uses a time scale without leap seconds, so that there are no hiccups in the equations. GPS time was set to UTC when GPS started in early 1980, when TAI-UTC was 19 s. Now that TAI-UTC is 32 s GPS system time is 13 s ahead of UTC.
John Chambers